Training model using annotated image

Hi, I am working on emotion recognition project where I am using EMOTIC dataset for training my model. All these images are annotated and all the info are stored in torch (.t7) file e.g “DiscreteContinuousAnnotations26_train.t7” for training images. In these images required person is annotated by bounding box whose emotions model needs to capture and learn gradually.
But I am facing issue where I don’t know how to read this torch file and also I am not getting how to feed this annotation file to model so that it will capture required person in the bounding box and will learn its expression.

Please help as I am doing this end to end project first time and I am quite beginner in this area.

I just want to know how to train model with images having bounding box (as label) around particular person so that it can learn only features of that person instead of entire image containing other objects and persons.